下载文中数据、代码、绘图结果

关于数据

如果想知道本文的关联规则数据是怎么来的，请阅读这篇文章

绘图函数

Python中似乎没有很方便的绘制网络图的函数。

下面是本人自行实现的绘图函数，如果想要运行，请点击上文的链接，下载数据和代码。

传入一个关联规则数据的DataFrame，这个DataFrame应该包含三列数据：antecedents，consequents，confidence，分别代表前件，后件，置信度。

def plot_rules_net(rules: pd.DataFrame):
    import matplotlib.patches as patches

    # 假设你有一个包含所有药材的列表
    items = list(
        set([item for sublist in rules['antecedents'].tolist() + rules['consequents'].tolist() for item in sublist]))

    # 计算药材数量，确定顶点数
    n_items = len(items)

    # 创建一个正n_items边形的顶点坐标
    radius = 5  # 可以调整半径
    angle = np.linspace(0, 2 * np.pi, n_items, endpoint=False)
    x = radius * np.cos(angle)
    y = radius * np.sin(angle)

    # 绘制正多边形和顶点
    fig, ax = plt.subplots(figsize=(10, 10))

    polygon = patches.RegularPolygon((0, 0), n_items, radius=radius, fill=False, edgecolor='k')
    ax.add_patch(polygon)

    def get_label_position(angle):
        label_offset_value = 0.2  # 定义一个变量来存储偏移量

        # 根据角度确定文本标签的对齐方式和位置
        if angle < np.pi / 2:
            ha, va = "center", "bottom"
            offset = np.array([label_offset_value, label_offset_value])
        elif angle < np.pi:
            ha, va = "center", "bottom"
            offset = np.array([-label_offset_value, label_offset_value])
        elif angle < 3 * np.pi / 2:
            ha, va = "center", "top"
            offset = np.array([-label_offset_value, -label_offset_value])
        else:
            ha, va = "center", "top"
            offset = np.array([label_offset_value, -label_offset_value])
        return ha, va, offset

    # 在绘制顶点的循环中调整文本位置
    for (i, j), label, angle in zip(zip(x, y), items, angle):
        ha, va, offset = get_label_position(angle)
        ax.plot(i, j, 'o', markersize=10)
        ax.text(i + offset[0], j + offset[1], label, fontsize=12, ha=ha, va=va)

    # 获取confidence的最小值和最大值
    min_confidence = rules['confidence'].min()
    max_confidence = rules['confidence'].max()
    # 使用colormap - 可以根据需要选择合适的colormap
    # 这里我们使用'Greens'，因为你想要的是颜色越深表示权重越大
    cmap = plt.get_cmap('Greens')

    # 线性映射函数，将confidence值映射到0-1之间，用于colormap
    def get_color(confidence):
        return cmap((confidence - min_confidence) / (max_confidence - min_confidence))

    # 绘制边
    for _, row in rules.iterrows():
        antecedents = row['antecedents']
        consequents = row['consequents']
        confidence = row['confidence']

        for antecedent in antecedents:
            for consequent in consequents:
                start_idx = items.index(antecedent)
                end_idx = items.index(consequent)

                start_point = (x[start_idx], y[start_idx])
                end_point = (x[end_idx], y[end_idx])

                color = get_color(confidence)

                # 修改箭头的绘制方式，使其从节点边缘出发
                ax.annotate("",
                            xy=end_point, xytext=start_point,
                            arrowprops=dict(arrowstyle="->", color=color,
                                            shrinkA=5, shrinkB=5,  # shrinkA和shrinkB应该是半径的大小，不是索引
                                            connectionstyle="arc3"),
                            )

    ax.set_xlim([-radius * 1.1, radius * 1.1])
    ax.set_ylim([-radius * 1.1, radius * 1.1])
    ax.axis('off')  # 隐藏坐标轴

    plt.suptitle('前24个最高频次药物的关联规则图', fontsize=20)  # 主标题
    plt.xlabel('颜色深代表置信度高', fontsize=14)  # X轴标签

    save_path = os.path.join('.', '关联规则网络图.jpg')

    plt.savefig(save_path)
    plt.show()

完整可运行的代码

下方就是完整可运行的代码，在本文的下载链接中也一并包含，如有需要请复制并运行。

import os

import matplotlib
import numpy as np
import pandas as pd
from matplotlib import pyplot as plt

plt.rcParams['font.sans-serif'] = ['Simhei']  # 显示中文标签
plt.rcParams['axes.unicode_minus'] = False

def plot_rules_net(rules: pd.DataFrame):
    import matplotlib.patches as patches

    # 假设你有一个包含所有药材的列表
    items = list(
        set([item for sublist in rules['antecedents'].tolist() + rules['consequents'].tolist() for item in sublist]))

    # 计算药材数量，确定顶点数
    n_items = len(items)

    # 创建一个正n_items边形的顶点坐标
    radius = 5  # 可以调整半径
    angle = np.linspace(0, 2 * np.pi, n_items, endpoint=False)
    x = radius * np.cos(angle)
    y = radius * np.sin(angle)

    # 绘制正多边形和顶点
    fig, ax = plt.subplots(figsize=(10, 10))

    polygon = patches.RegularPolygon((0, 0), n_items, radius=radius, fill=False, edgecolor='k')
    ax.add_patch(polygon)

    def get_label_position(angle):
        label_offset_value = 0.2  # 定义一个变量来存储偏移量

        # 根据角度确定文本标签的对齐方式和位置
        if angle < np.pi / 2:
            ha, va = "center", "bottom"
            offset = np.array([label_offset_value, label_offset_value])
        elif angle < np.pi:
            ha, va = "center", "bottom"
            offset = np.array([-label_offset_value, label_offset_value])
        elif angle < 3 * np.pi / 2:
            ha, va = "center", "top"
            offset = np.array([-label_offset_value, -label_offset_value])
        else:
            ha, va = "center", "top"
            offset = np.array([label_offset_value, -label_offset_value])
        return ha, va, offset

    # 在绘制顶点的循环中调整文本位置
    for (i, j), label, angle in zip(zip(x, y), items, angle):
        ha, va, offset = get_label_position(angle)
        ax.plot(i, j, 'o', markersize=10)
        ax.text(i + offset[0], j + offset[1], label, fontsize=12, ha=ha, va=va)

    # 获取confidence的最小值和最大值
    min_confidence = rules['confidence'].min()
    max_confidence = rules['confidence'].max()
    # 使用colormap - 可以根据需要选择合适的colormap
    # 这里我们使用'Greens'，因为你想要的是颜色越深表示权重越大
    cmap = plt.get_cmap('Greens')

    # 线性映射函数，将confidence值映射到0-1之间，用于colormap
    def get_color(confidence):
        return cmap((confidence - min_confidence) / (max_confidence - min_confidence))

    # 绘制边
    for _, row in rules.iterrows():
        antecedents = row['antecedents']
        consequents = row['consequents']
        confidence = row['confidence']

        for antecedent in antecedents:
            for consequent in consequents:
                start_idx = items.index(antecedent)
                end_idx = items.index(consequent)

                start_point = (x[start_idx], y[start_idx])
                end_point = (x[end_idx], y[end_idx])

                color = get_color(confidence)

                # 修改箭头的绘制方式，使其从节点边缘出发
                ax.annotate("",
                            xy=end_point, xytext=start_point,
                            arrowprops=dict(arrowstyle="->", color=color,
                                            shrinkA=5, shrinkB=5,  # shrinkA和shrinkB应该是半径的大小，不是索引
                                            connectionstyle="arc3"),
                            )

    ax.set_xlim([-radius * 1.1, radius * 1.1])
    ax.set_ylim([-radius * 1.1, radius * 1.1])
    ax.axis('off')  # 隐藏坐标轴

    plt.suptitle('前24个最高频次药物的关联规则图', fontsize=20)  # 主标题
    plt.xlabel('颜色深代表置信度高', fontsize=14)  # X轴标签

    save_path = os.path.join('.', '关联规则网络图.jpg')

    plt.savefig(save_path)
    plt.show()



freq = pd.read_excel(r'万条处方的药物出现频次.xlsx')

fd = {k : v for _, (k, v) in freq.iterrows()}

# 指定保留前24最高频次的中药材
most_freq_num = 24
top_24_herbs = sorted(fd, key=lambda x: fd.get(x), reverse=True)[:most_freq_num]

# 读取关联规则分析的结果
rules = pd.read_excel('关联规则分析结果.xlsx')
rules['antecedents'] = rules['antecedents'].apply(lambda x: x.split(', '))
rules['consequents'] = rules['consequents'].apply(lambda x: x.split(', '))

# 过滤关联规则，仅保留包含这24种药物的规则
filtered_rules = rules[rules['antecedents'].apply(lambda x: any(item in x for item in top_24_herbs)) &
                       rules['consequents'].apply(lambda x: any(item in x for item in top_24_herbs))]

plot_rules_net(filtered_rules)

运行结果

颜色越粗，置信度越高。
我们可以看到砂仁和白芍，还有荷叶和连翘等等，有着很高的置信度。